CN-121980197-A - Tunnel structure bad loss prediction method and system based on multi-source data fusion awareness

CN121980197ACN 121980197 ACN121980197 ACN 121980197ACN-121980197-A

Abstract

The invention provides a tunnel structure bad loss prediction method and system based on multi-source data fusion perception, which relate to the technical field of computers and comprise the steps of obtaining original test data generated by a cold region heavy haul railway tunnel structure bad loss test system; extracting features according to the original test data to obtain feature vector sets representing a dynamic state, a static bearing state and an apparent damage state of the structure respectively, carrying out feature fusion according to the feature vector sets and environmental parameter time sequence data to obtain advanced feature representation, carrying out feature screening and dimension reduction according to the advanced feature representation to obtain an optimal feature subset after dimension reduction, carrying out model mapping according to the optimal feature subset to construct a bad damage prediction model, and inputting field original monitoring data into the bad damage prediction model to obtain a prediction result. The invention obviously improves the depth of state perception, the reliability of assessment and early warning and the scientificity of maintenance decision.

Inventors

YANG WENBO
YANG LINLIN
QIU ZHIJIE
LI HAOYU
WU BINGYIN
YAO CHAOFAN

Assignees

西南交通大学

Dates

Publication Date: 20260505
Application Date: 20260403

Claims (10)

1.A tunnel structure bad loss prediction method based on multi-source data fusion awareness is characterized by comprising the following steps: The method comprises the steps of obtaining original test data generated by a cold region heavy haul railway tunnel structure bad damage test system, wherein the original test data comprise vibration response time sequence data, internal force time sequence data, apparent image data and environmental parameter time sequence data of a lining test piece under the coupling effect of freeze thawing cycle and heavy haul load; extracting features according to the original test data to respectively obtain feature vector sets representing a dynamic state, a static bearing state and an apparent damage state of the structure; Performing feature fusion according to the feature vector set and the environmental parameter time sequence data, adaptively distributing weights for the features of different sources by combining the environmental states, combining the weighted features, and performing nonlinear interaction to obtain advanced feature representation; performing feature screening and dimension reduction according to the advanced feature representation, and obtaining an optimal feature subset after dimension reduction by calculating feature importance and arranging the feature importance in a descending order of importance; Performing model mapping according to the optimal feature subset, and constructing and obtaining an inferior loss prediction model by training a nonlinear regression function based on the optimal feature subset and a known inferior loss state label; and inputting the field original monitoring data into the bad damage prediction model to obtain a prediction result of the future safety state of the tunnel structure.
2. The method for predicting the inferior loss of the tunnel structure based on multi-source data fusion awareness of claim 1, wherein the feature extraction is performed according to the original test data to obtain feature vector sets for representing a dynamic state, a bearing state and an apparent damage state of the structure respectively, and the method comprises the following steps: Performing dynamic characteristic extraction processing according to the vibration response time sequence data, performing wavelet packet decomposition on the vibration response time sequence data to obtain signal components of different frequency bands, and calculating energy of each frequency band, amplitude factors and peak factors of the signals to obtain characteristic vectors representing structural impact characteristics; carrying out bearing characteristic extraction processing according to the internal force time sequence data, and extracting local and deep patterns from original stress or strain waveforms by constructing a one-dimensional convolutional neural network to obtain a characteristic vector representing the bearing state of the structure; And performing visual feature extraction processing according to the apparent image data, and quantifying the total length, average width, maximum width and fractal dimension of the crack by performing pixel-level crack segmentation, skeleton extraction and width measurement along the normal direction of the skeleton on the image to obtain a feature vector representing the apparent damage state.
3. The method for predicting the inferior loss of the tunnel structure based on multi-source data fusion awareness of claim 1, wherein the feature fusion is performed according to the feature vector set and the environmental parameter time sequence data, and advanced feature representation is obtained by adaptively distributing weights to features of different sources in combination with environmental states and combining the weighted features to perform nonlinear interaction, comprising: Carrying out standardization processing according to the feature vector set, and calculating the mean value and standard deviation of each feature vector to enable all features to be in the same comparable scale, so as to obtain a standardized feature vector; Performing self-adaptive weighting processing according to the normalized feature vector and the environmental parameter time sequence data, splicing the environmental feature vector quantized with the freeze thawing cycle times, the negative temperature duration and the temperature gradient with each data source feature vector, generating a weight vector related to the current freeze thawing environmental state by using an attention mechanism, and scaling the feature channel to obtain a re-weighted feature vector; Performing splicing processing according to all the re-weighted feature vectors, and connecting the feature vectors of different data sources in the feature dimension to obtain a spliced fusion feature vector; And carrying out deep nonlinear interaction processing according to the fusion feature vector, inputting the fusion feature vector into a fully-connected neural network to carry out nonlinear transformation and feature combination, and mining a cross-modal damage association mode under the coupling action of freeze thawing and load to obtain the advanced feature representation.
4. The method for predicting the inferior loss of the tunnel structure based on multi-source data fusion awareness according to claim 1, wherein the method for predicting the inferior loss of the tunnel structure based on multi-source data fusion awareness is characterized by performing feature screening and dimension reduction according to the advanced feature representation, obtaining an optimal feature subset after dimension reduction by calculating feature importance and arranging the feature importance in a descending order of importance, and comprises the following steps: Model training is carried out according to the advanced feature representation and the corresponding known bad damage state label, and the contribution degree of each feature in the advanced feature representation to the tunnel structure bad damage state discrimination is evaluated based on the reduction amount of the base non-purity by training a random forest model to obtain the importance score of each feature; Performing feature sorting according to the importance scores, and arranging all features according to the importance scores from high to low to obtain a feature list with descending importance arrangement; and carrying out feature subset selection processing according to the feature list, and obtaining an optimal feature subset by selecting the top K features with highest importance ranking from the feature list.
5. The method for predicting the inferior loss of the tunnel structure based on multi-source data fusion awareness according to claim 1, wherein the method for constructing and obtaining the prediction model of the inferior loss by training a nonlinear regression function based on the optimal feature subset and the known inferior loss state label is characterized by performing model mapping according to the optimal feature subset and comprises the following steps: Performing iterative model training treatment according to the optimal feature subset and the corresponding known bad loss state label, adopting a gradient lifting decision tree algorithm, fitting a new regression tree by taking the prediction residual of the previous round of model as a target, and accumulating the prediction results of the gradient lifting decision tree round by round to minimize the whole loss function to obtain a preliminary regression model; performing numerical range constraint processing according to the original output of the preliminary regression model, and mapping a continuous predicted value into a range from 0 to 1 by applying a Sigmoid function to obtain a continuous inferior loss index; And carrying out engineering state mapping processing according to the continuous deterioration index, and constructing and obtaining a deterioration prediction model by equally dividing a numerical interval into four subintervals and respectively endowing each subinterval with a security level and engineering state description.
6. A tunnel structure inferior loss prediction system based on multi-source data fusion awareness is characterized by comprising: The acquisition module is used for acquiring original test data generated by the cold region heavy haul railway tunnel structure bad damage test system, wherein the original test data comprises vibration response time sequence data, internal force time sequence data, apparent image data and environmental parameter time sequence data of the lining test piece under the coupling action of freezing and thawing cycle and heavy haul load; the extraction module is used for extracting the characteristics according to the original test data to respectively obtain characteristic vector sets representing the dynamic state, the static bearing state and the apparent damage state of the structure; The fusion module is used for carrying out feature fusion according to the feature vector set and the environment parameter time sequence data, and obtaining advanced feature representation by self-adaptively distributing weights for the features of different sources by combining the environment states and combining the weighted features to carry out nonlinear interaction; the screening module is used for carrying out feature screening and dimension reduction according to the advanced feature representation, and obtaining an optimal feature subset after dimension reduction by calculating feature importance and arranging the feature importance in a descending order of importance; The construction module is used for carrying out model mapping according to the optimal feature subset, and constructing and obtaining an inferior loss prediction model by training a nonlinear regression function based on the optimal feature subset and a known inferior loss state label; And the prediction module is used for inputting the field original monitoring data into the bad damage prediction model to obtain a prediction result of the future safety state of the tunnel structure.
7. The system for predicting deterioration of a tunnel structure based on multi-source data fusion awareness of claim 6, wherein the extracting module comprises: The first extraction unit is used for carrying out dynamic characteristic extraction processing according to the vibration response time sequence data, carrying out wavelet packet decomposition on the vibration response time sequence data to obtain signal components of different frequency bands, and calculating energy of each frequency band, amplitude factors and peak factors of the signals to obtain characteristic vectors representing impact characteristics of the structure; The second extraction unit is used for carrying out bearing characteristic extraction processing according to the internal force time sequence data, and extracting local and deep patterns from original stress or strain waveforms by constructing a one-dimensional convolutional neural network to obtain a characteristic vector representing the bearing state of the structure; And the third extraction unit is used for carrying out visual feature extraction processing according to the apparent image data, and obtaining a feature vector representing the apparent damage state by carrying out pixel-level crack segmentation, skeleton extraction and width measurement along the normal direction of the skeleton on the image and quantifying the total length, average width, maximum width and fractal dimension of the crack.
8. The system for predicting deterioration of a tunnel structure based on multi-source data fusion awareness of claim 6, wherein the fusion module comprises: The first fusion unit is used for carrying out standardization processing according to the feature vector set, and all features are in the same comparable scale by calculating the mean value and standard deviation of each feature vector to obtain a standardized feature vector; the second fusion unit is used for carrying out self-adaptive weighting processing according to the normalized feature vector and the environment parameter time sequence data, splicing the environment feature vector quantized with the freeze thawing cycle times, the negative temperature duration time and the temperature gradient with each data source feature vector, generating a weight vector related to the current freeze thawing environment state by using an attention mechanism, and scaling the feature channel to obtain a re-weighted feature vector; the third fusion unit is used for carrying out splicing processing according to all the re-weighted feature vectors, and connecting the feature vectors of different data sources in the feature dimension to obtain a spliced fusion feature vector; And the fourth fusion unit is used for carrying out deep nonlinear interaction processing according to the fusion feature vector, inputting the fusion feature vector into a fully-connected neural network to carry out nonlinear transformation and feature combination, and excavating a cross-modal damage association mode under the coupling action of freeze thawing and load to obtain the advanced feature representation.
9. The system for predicting deterioration of a tunnel structure based on multi-source data fusion awareness of claim 6, wherein the screening module comprises: the first screening unit is used for carrying out model training treatment according to the advanced feature representation and the corresponding known bad loss state label, and obtaining importance scores of each feature by training a random forest model and evaluating contribution degree of each feature in the advanced feature representation to the discrimination of the tunnel structure bad loss state based on the reduction amount of the base non-purity; The second screening unit is used for carrying out feature sorting processing according to the importance scores, and obtaining a feature list with the importance arranged in descending order by arranging all the features according to the importance scores from high to low; And the third screening unit is used for carrying out feature subset selection processing according to the feature list, and obtaining an optimal feature subset by selecting the top K features with highest importance ranking from the feature list.
10. The system for predicting deterioration of a tunnel structure based on multi-source data fusion awareness of claim 6, wherein the building module comprises: the first construction unit is used for carrying out iterative model training treatment according to the optimal feature subset and the corresponding known bad loss state label, fitting a new regression tree by taking the prediction residual of the previous round of model as a target through adopting a gradient lifting decision tree algorithm, and accumulating the prediction results of the gradient lifting decision tree round by round to minimize the whole loss function so as to obtain a preliminary regression model; The second construction unit is used for carrying out numerical range constraint processing according to the original output of the preliminary regression model, and mapping the continuous predicted value into a range from 0 to 1 by applying a Sigmoid function to obtain a continuous inferior loss index; And the third construction unit is used for carrying out engineering state mapping processing according to the continuous inferior loss index, and constructing and obtaining an inferior loss prediction model by equally dividing a numerical interval into four subintervals and respectively endowing each subinterval with a security level and engineering state description.

Description

Tunnel structure bad loss prediction method and system based on multi-source data fusion awareness Technical Field The invention relates to the technical field of computers, in particular to a tunnel structure bad loss prediction method and system based on multi-source data fusion awareness. Background Along with the continuous extension of the railway network of China to the complex environments such as high cold, high altitude and the like, the long-term service safety of the high cold heavy-load railway tunnel faces a serious challenge. Under the coupling effect of continuous freeze thawing cycle and heavy load train dynamic load, tunnel lining structure is liable to generate various recessive and dominant damages, and its evolution mechanism is complicated, and traditional monitoring means mainly comprising single sensors such as manual inspection, crack meter, strain gauge, etc. have the problems of low efficiency, high risk, incomplete coverage, etc. under severe environment, and it is difficult to comprehensively capture the complete information of structural damage. The existing automatic monitoring technology can acquire time sequence data of a specific section for a long time, but the time sequence data are distributed in a punctiform and discrete space mode, so that the time-space discontinuity of the data is caused, a data field reflecting the continuous evolution of the state of the full-line structure of the tunnel cannot be constructed, and the integral evaluation of the spatial distribution and evolution rule of the damage is restricted. In the data analysis level, the existing method is mostly dependent on an empirical formula and shallow statistical analysis, is difficult to effectively process massive heterogeneous multi-source monitoring data, cannot automatically mine deep damage evolution modes and sensitive features under the driving of multi-factor coupling such as freeze thawing, loading and the like, and causes insufficient precision and reliability of state evaluation. More importantly, the current technical system focuses on diagnosis of history and current state, and lacks an effective model capable of scientifically predicting future degradation trend and safe state of the structure, so that maintenance decision is seriously lagged, and the system falls into a passive situation of post-treatment. Therefore, how to break through the technical bottlenecks of single sensing means, data islanding, analysis shallowing, prediction capability deficiency and the like, realize the transition from discrete point detection, passive response to continuous sensing, intelligent evaluation and pre-warning, and become a core technical problem to be solved in the field of security assurance of high and cold heavy load railway tunnels. Based on the shortcomings of the prior art, a method and a system for predicting the inferior loss of a tunnel structure based on multi-source data fusion perception are needed. Disclosure of Invention The invention aims to provide a tunnel structure bad loss prediction method and system based on multi-source data fusion awareness so as to solve the problems. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: In a first aspect, the present application provides a tunnel structure bad loss prediction method based on multi-source data fusion awareness, including: The method comprises the steps of obtaining original test data generated by a cold region heavy haul railway tunnel structure bad damage test system, wherein the original test data comprise vibration response time sequence data, internal force time sequence data, apparent image data and environmental parameter time sequence data of a lining test piece under the coupling effect of freeze thawing cycle and heavy haul load; extracting features according to the original test data to respectively obtain feature vector sets representing a dynamic state, a static bearing state and an apparent damage state of the structure; Performing feature fusion according to the feature vector set and the environmental parameter time sequence data, adaptively distributing weights for the features of different sources by combining the environmental states, combining the weighted features, and performing nonlinear interaction to obtain advanced feature representation; performing feature screening and dimension reduction according to the advanced feature representation, and obtaining an optimal feature subset after dimension reduction by calculating feature importance and arranging the feature importance in a descending order of importance; Performing model mapping according to the optimal feature subset, and constructing and obtaining an inferior loss prediction model by training a nonlinear regression function based on the optimal feature subset and a known inferior loss state label; and inputting the field original monitoring data into the bad damage prediction model to obtain a p